net/mlx5e: Add cache for HW modify header IDs
authorOr Gerlitz <ogerlitz@mellanox.com>
Thu, 4 May 2017 18:46:11 +0000 (21:46 +0300)
committerSaeed Mahameed <saeedm@mellanox.com>
Thu, 8 Jun 2017 11:12:00 +0000 (14:12 +0300)
Packets belonging to flows which are different by matching may still need
to go through the same header re-write. Add a cache for header re-write IDs
keyed by the binary chain of modify header actions.

The caching is supported for both eswitch and NIC use-cases, where the
actual conversion of the code to use caching comes in next patches, one
per use-case.

Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Reviewed-by: Paul Blakey <paulb@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
drivers/net/ethernet/mellanox/mlx5/core/en.h
drivers/net/ethernet/mellanox/mlx5/core/en_tc.c
drivers/net/ethernet/mellanox/mlx5/core/eswitch.c
drivers/net/ethernet/mellanox/mlx5/core/eswitch.h

index 2fd044b238750fedc61384fda55bf25206d25b51..f4b95dbd1c7f2304a87315890259865769237799 100644 (file)
@@ -623,6 +623,8 @@ struct mlx5e_tc_table {
 
        struct rhashtable_params        ht_params;
        struct rhashtable               ht;
+
+       DECLARE_HASHTABLE(mod_hdr_tbl, 8);
 };
 
 struct mlx5e_vlan_table {
index a9feddc31667ca5f6821dc615cc8519a199b923b..e1217c2279ab1bf02868e7087c1deed9925bdcca 100644 (file)
@@ -69,7 +69,8 @@ struct mlx5e_tc_flow {
        u64                     cookie;
        u8                      flags;
        struct mlx5_flow_handle *rule;
-       struct list_head        encap; /* flows sharing the same encap */
+       struct list_head        encap;   /* flows sharing the same encap ID */
+       struct list_head        mod_hdr; /* flows sharing the same mod hdr ID */
        union {
                struct mlx5_esw_flow_attr esw_attr[0];
                struct mlx5_nic_flow_attr nic_attr[0];
@@ -90,6 +91,135 @@ enum {
 #define MLX5E_TC_TABLE_NUM_ENTRIES 1024
 #define MLX5E_TC_TABLE_NUM_GROUPS 4
 
+struct mod_hdr_key {
+       int num_actions;
+       void *actions;
+};
+
+struct mlx5e_mod_hdr_entry {
+       /* a node of a hash table which keeps all the mod_hdr entries */
+       struct hlist_node mod_hdr_hlist;
+
+       /* flows sharing the same mod_hdr entry */
+       struct list_head flows;
+
+       struct mod_hdr_key key;
+
+       u32 mod_hdr_id;
+};
+
+#define MLX5_MH_ACT_SZ MLX5_UN_SZ_BYTES(set_action_in_add_action_in_auto)
+
+static inline u32 hash_mod_hdr_info(struct mod_hdr_key *key)
+{
+       return jhash(key->actions,
+                    key->num_actions * MLX5_MH_ACT_SZ, 0);
+}
+
+static inline int cmp_mod_hdr_info(struct mod_hdr_key *a,
+                                  struct mod_hdr_key *b)
+{
+       if (a->num_actions != b->num_actions)
+               return 1;
+
+       return memcmp(a->actions, b->actions, a->num_actions * MLX5_MH_ACT_SZ);
+}
+
+static int mlx5e_attach_mod_hdr(struct mlx5e_priv *priv,
+                               struct mlx5e_tc_flow *flow,
+                               struct mlx5e_tc_flow_parse_attr *parse_attr)
+{
+       struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
+       int num_actions, actions_size, namespace, err;
+       struct mlx5e_mod_hdr_entry *mh;
+       struct mod_hdr_key key;
+       bool found = false;
+       u32 hash_key;
+
+       num_actions  = parse_attr->num_mod_hdr_actions;
+       actions_size = MLX5_MH_ACT_SZ * num_actions;
+
+       key.actions = parse_attr->mod_hdr_actions;
+       key.num_actions = num_actions;
+
+       hash_key = hash_mod_hdr_info(&key);
+
+       if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
+               namespace = MLX5_FLOW_NAMESPACE_FDB;
+               hash_for_each_possible(esw->offloads.mod_hdr_tbl, mh,
+                                      mod_hdr_hlist, hash_key) {
+                       if (!cmp_mod_hdr_info(&mh->key, &key)) {
+                               found = true;
+                               break;
+                       }
+               }
+       } else {
+               namespace = MLX5_FLOW_NAMESPACE_KERNEL;
+               hash_for_each_possible(priv->fs.tc.mod_hdr_tbl, mh,
+                                      mod_hdr_hlist, hash_key) {
+                       if (!cmp_mod_hdr_info(&mh->key, &key)) {
+                               found = true;
+                               break;
+                       }
+               }
+       }
+
+       if (found)
+               goto attach_flow;
+
+       mh = kzalloc(sizeof(*mh) + actions_size, GFP_KERNEL);
+       if (!mh)
+               return -ENOMEM;
+
+       mh->key.actions = (void *)mh + sizeof(*mh);
+       memcpy(mh->key.actions, key.actions, actions_size);
+       mh->key.num_actions = num_actions;
+       INIT_LIST_HEAD(&mh->flows);
+
+       err = mlx5_modify_header_alloc(priv->mdev, namespace,
+                                      mh->key.num_actions,
+                                      mh->key.actions,
+                                      &mh->mod_hdr_id);
+       if (err)
+               goto out_err;
+
+       if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
+               hash_add(esw->offloads.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
+       else
+               hash_add(priv->fs.tc.mod_hdr_tbl, &mh->mod_hdr_hlist, hash_key);
+
+attach_flow:
+       list_add(&flow->mod_hdr, &mh->flows);
+       if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
+               flow->esw_attr->mod_hdr_id = mh->mod_hdr_id;
+       else
+               flow->nic_attr->mod_hdr_id = mh->mod_hdr_id;
+
+       return 0;
+
+out_err:
+       kfree(mh);
+       return err;
+}
+
+static void mlx5e_detach_mod_hdr(struct mlx5e_priv *priv,
+                                struct mlx5e_tc_flow *flow)
+{
+       struct list_head *next = flow->mod_hdr.next;
+
+       list_del(&flow->mod_hdr);
+
+       if (list_empty(next)) {
+               struct mlx5e_mod_hdr_entry *mh;
+
+               mh = list_entry(next, struct mlx5e_mod_hdr_entry, flows);
+
+               mlx5_modify_header_dealloc(priv->mdev, mh->mod_hdr_id);
+               hash_del(&mh->mod_hdr_hlist);
+               kfree(mh);
+       }
+}
+
 static struct mlx5_flow_handle *
 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
                      struct mlx5e_tc_flow_parse_attr *parse_attr,
@@ -1939,6 +2069,8 @@ int mlx5e_tc_init(struct mlx5e_priv *priv)
 {
        struct mlx5e_tc_table *tc = &priv->fs.tc;
 
+       hash_init(tc->mod_hdr_tbl);
+
        tc->ht_params = mlx5e_tc_flow_ht_params;
        return rhashtable_init(&tc->ht, &tc->ht_params);
 }
index 81dfcd90b1f54b5430518080040801a63c63278e..37927156f258cb50cb29f0ea4d23a559b4f8a81f 100644 (file)
@@ -1769,6 +1769,7 @@ int mlx5_eswitch_init(struct mlx5_core_dev *dev)
        }
 
        hash_init(esw->offloads.encap_tbl);
+       hash_init(esw->offloads.mod_hdr_tbl);
        mutex_init(&esw->state_lock);
 
        for (vport_num = 0; vport_num < total_vports; vport_num++) {
index b746f62c8c79b17a64c0923665917170a075092a..834a33050969719ca6e13b27c11b6eb895c02257 100644 (file)
@@ -207,6 +207,7 @@ struct mlx5_esw_offload {
        struct mlx5_flow_group *vport_rx_group;
        struct mlx5_eswitch_rep *vport_reps;
        DECLARE_HASHTABLE(encap_tbl, 8);
+       DECLARE_HASHTABLE(mod_hdr_tbl, 8);
        u8 inline_mode;
        u64 num_flows;
        u8 encap;